Fault diagnosis in PV systems is studied broadly from various aspects. There are different techniques for FFD in PV systems. Among them unsupervised methods are appreciated for not needing labeled datasets or huge amounts of historical data. Clustering the data to study the similarities and differences is one of reliable unsupervised methods. To obtain proper accuracy in this paper, we introduce new integrated techniques for fault diagnosis and prediction. By generating different PV fault scenarios, different clustering techniques yield labels separately and all these clustering labels are gathered in an ensemble area to obtain best labels by voting. Voting majority empowers our ensemble clustering techniques. By interpreting the centroids of consensus clusters based on electrical characteristics of faults, we can generate fault labels. The final classification is outlined by neural networks (MLP). The results and experiments in a real test PV system show higher accuracy than other state of art unsupervised techniques.